AntonioJun's picture
Add files using upload-large-folder tool
8e80498 verified
Raw
History Blame Contribute Delete
6.63 kB
"""Build hypothesis spatial codes from existing native or combined geometry caches."""
from __future__ import annotations
import argparse
import gzip
import json
import os
from pathlib import Path
import pickle
import sys
import types
WORKSPACE_ROOT = Path(__file__).resolve().parent.parent
if str(WORKSPACE_ROOT) not in sys.path:
sys.path.insert(0, str(WORKSPACE_ROOT))
from encoder import adapters # noqa: E402
from encoder import config as encoder_config # noqa: E402
from experiments import config, loader # noqa: E402
from experiments import adapters as format_adapters # noqa: E402
class CachedDA3Prediction:
"""Attribute container used only to deserialize cached DA3 Prediction objects."""
class CachedAddictDict(dict):
"""Minimal attribute-access dictionary required by cached DA3 metadata."""
def __getattr__(self, name):
if name.startswith("__") and name.endswith("__"):
raise AttributeError(name)
try:
return self[name]
except KeyError as exc:
raise AttributeError(name) from exc
def __setattr__(self, name, value):
self[name] = value
def install_da3_pickle_compatibility() -> None:
"""Provide the exact pickle class when the original DA3 package is unavailable."""
try:
__import__("depth_anything_3.specs")
except ModuleNotFoundError:
package = types.ModuleType("depth_anything_3")
package.__path__ = []
specs = types.ModuleType("depth_anything_3.specs")
CachedDA3Prediction.__module__ = "depth_anything_3.specs"
CachedDA3Prediction.__name__ = "Prediction"
CachedDA3Prediction.__qualname__ = "Prediction"
specs.Prediction = CachedDA3Prediction
package.specs = specs
sys.modules["depth_anything_3"] = package
sys.modules["depth_anything_3.specs"] = specs
try:
__import__("addict.addict")
except ModuleNotFoundError:
package = types.ModuleType("addict")
package.__path__ = []
module = types.ModuleType("addict.addict")
CachedAddictDict.__module__ = "addict.addict"
CachedAddictDict.__name__ = "Dict"
CachedAddictDict.__qualname__ = "Dict"
module.Dict = CachedAddictDict
package.addict = module
package.Dict = CachedAddictDict
sys.modules["addict"] = package
sys.modules["addict.addict"] = module
def configure_single_scene_threads() -> int:
"""Give a direct single-scene run every CPU available to the process."""
try:
count = max(1, len(os.sched_getaffinity(0)))
except AttributeError:
count = max(1, os.cpu_count() or 1)
value = str(count)
for variable in (
"OMP_NUM_THREADS",
"MKL_NUM_THREADS",
"OPENBLAS_NUM_THREADS",
"NUMEXPR_NUM_THREADS",
"VSI_KD_WORKERS",
):
os.environ.setdefault(variable, value)
return count
def load_existing_geometry(scene, depth, input_selection, tracking, frame_count):
"""Read existing caches without ever creating or modifying source caches."""
combined_path = Path(
encoder_config.cache_file(scene, depth, input_selection, tracking, frame_count)
)
if combined_path.is_file():
with gzip.open(combined_path, "rb") as stream:
return adapters.validate(pickle.load(stream))
da3_path = Path(
encoder_config.da3_cache_file(scene, depth, input_selection, frame_count)
)
sam3_path = Path(
encoder_config.sam3_cache_file(scene, input_selection, tracking, frame_count)
)
missing = [str(path) for path in (da3_path, sam3_path) if not path.is_file()]
if missing:
raise FileNotFoundError(
"required native cache(s) are missing: " + ", ".join(missing)
)
install_da3_pickle_compatibility()
geometry = adapters.adapt(
encoder_config.MODEL,
scene=scene,
root=str(encoder_config.CACHE_ROOT),
rebuild=False,
da3_path=str(da3_path),
sam3_path=str(sam3_path),
)
return adapters.validate(geometry)
def run_scene(
scene,
hypothesis,
depth="metric",
tracking="tracking",
input_selection="uniform",
frame_count=64,
rebuild=False,
spatial_code_format="explicit",
):
output = config.spatial_code_path(
scene,
hypothesis,
depth,
tracking,
input_selection,
frame_count,
spatial_code_format,
)
if output.is_file() and not rebuild:
with output.open(encoding="utf-8") as stream:
return json.load(stream), "loaded", output
geometry = load_existing_geometry(
scene, depth, input_selection, tracking, frame_count
)
geometry_math = loader.load_hypothesis(hypothesis)
code, *_ = format_adapters.build(geometry_math, geometry, spatial_code_format)
output.parent.mkdir(parents=True, exist_ok=True)
geometry_math.dump_spatial_code(code, str(output))
return code, "built", output
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("scene", nargs="?")
parser.add_argument("--hypothesis")
parser.add_argument(
"--depth", default="metric", choices=encoder_config.DEPTH_VARIANTS
)
parser.add_argument(
"--input",
default="uniform",
choices=encoder_config.INPUT_SELECTIONS,
dest="input_selection",
)
parser.add_argument(
"--tracking", default="tracking", choices=encoder_config.TRACKING_MODES
)
parser.add_argument("--frames", type=int, default=64)
parser.add_argument(
"--format",
default="explicit",
choices=config.SPATIAL_CODE_FORMATS,
dest="spatial_code_format",
)
parser.add_argument("--rebuild", action="store_true")
parser.add_argument("--list", action="store_true", dest="list_only")
args = parser.parse_args()
if args.list_only:
print("\n".join(loader.list_hypotheses()))
return
if not args.scene:
parser.error("scene is required unless --list is used")
if not args.hypothesis:
parser.error("--hypothesis is required unless --list is used")
if args.frames < 1:
parser.error("--frames must be positive")
configure_single_scene_threads()
_, status, path = run_scene(
args.scene,
args.hypothesis,
args.depth,
args.tracking,
args.input_selection,
args.frames,
args.rebuild,
args.spatial_code_format,
)
print(f"[{args.scene}] {status} -> {path}")
if __name__ == "__main__":
main()